A healthy diet is paramount for physical and mental health( Reference Mozaffarian, Appel and Van Horn 1 , Reference O’Neil, Quirk and Housden 2 ), and improving population diets was declared a priority area of action at the United Nations High Level Meeting on Prevention and Control of Non-Communicable Diseases( Reference Beaglehole, Bonita and Horton 3 ). Diet quality depends on personal food choices, which are driven by food prices as well as by culture, taste and convenience( Reference French 4 ). Epidemiological evidence indicates that better diet quality is associated with higher diet costs( Reference Rao, Afshin and Singh 5 ). Furthermore, higher price indices for fruits and vegetables were linked to higher BMI in children aged 2–9 years( Reference Beydoun, Shroff and Chen 6 ).
From 2000 to 2010, diet costs increased disproportionately in European countries, with the greatest increases in South European countries such as Spain (31·2 %) compared with 17·2 % in Germany or 20·6 % in Sweden( 7 ). During that same decade, food prices rose more sharply in Spain for healthy food choices compared with less healthy foods( 8 ). The cost of foods low in energy density and rich in nutrients, such as fruits, increased by 51·0 %, whereas pastries or confectionery products, high in energy density but low in nutrient density, increased by 10·1 and 23·1 %, respectively. High-density energy consumption has been related to low nutrient adequacy( Reference Schroder, Covas and Elosua 9 , Reference Schroder, Vila and Marrugat 10 ), weight gain( Reference Savage, Marini and Birch 11 ) and risk of obesity( Reference Vernarelli, Mitchell and Rolls 12 ).
It is unknown how increases in individual diet cost, driven by rising food prices, affects consumers’ food choices and, consequently, overall diet quality. Therefore, the aim of the present study was to analyse the prospective association between changes in individual diet cost and changes in diet quality in a representative Spanish population. In addition, we determined the impact of changes in diet cost on body weight.
Data were obtained from a population-based survey conducted in Girona (Spain) in 2000 and 2009. The baseline survey examined a randomly selected, population-based sample of 3058 men and women aged 25–74 years (participation rate: 71·0 %). Of the 3058 participants in the baseline survey in 2000, 2715 non-institutionalised participants who still resided in the catchment area in 2009 were invited to participate in the follow-up study (online Supplementary Fig. S1), and 2181 of these individuals attended the re-examination in 2009–2010. This represents a 19·7 % loss to follow-up after 10 years, resulting in an acceptable follow-up rate of 80·3 %. Finally, 3·2 % (n 69) of the participants had missing dietary data at baseline or at follow-up and were excluded from the analysis. The final sample size included 2112 participants with complete follow-up data. Participants were duly informed and signed their consent to participate in the study. The project was approved by the local Ethics Committee (CEIC-PSMAR, Barcelona, Spain).
Food consumption was determined using a validated FFQ, administered by a trained interviewer at baseline and at follow-up( Reference Benitez-Arciniega, Mendez and Baena-Diez 13 , Reference Schroder, Covas and Marrugat 14 ). In a 166-item food list including alcoholic and non-alcoholic beverages, participants indicated their usual consumption and chose from ten frequency categories ranging from never or less than once per month to six or more times per d.
Monetary diet cost
Food prices were obtained from the food price database of the Spanish Ministry of Economy and Competitiveness( 8 ). The average prices for many food items (not including commercial fast foods) are updated every month in this database. For this study, we calculated food prices for 2000 and 2010, based on the average cumulated prices reported for each of those 2 years. Prices were not available for the following foods (2 %): paella, cannelloni and pizza. Prices for fast food items were obtained by a search of corporate websites. Individuals’ daily diet cost and the monetary diet cost per 8·36 MJ of energy intake/d (hereinafter, energy-adjusted diet cost) were calculated.
Measurement of diet quality
Diet quality was determined by adherence to the Mediterranean diet and by measuring the energy density of the daily diet. We chose these two indices of diet quality from among the numerous available indicators because of their good construct validity and established association with health outcomes( Reference Schroder, Covas and Elosua 9 – Reference Vernarelli, Mitchell and Rolls 12 , Reference Sofi, Macchi and Abbate 15 – Reference Schröder, Salas-Salvadó and Martínez-González 17 ).
Modified Mediterranean diet score recommendations
Assessing adherence to the Mediterranean diet by a score based on population-based food consumption distribution is, by definition, specific to a particular population, making it difficult to compare results between studies. To overcome the limitation regarding comparability of results, we calculated the modified Mediterranean diet score recommendations (MDS-rec) as previously described( Reference Funtikova, Benitez-Arciniega and Gomez 18 ). In brief, consumption that meets recommended intakes for Spanish adults of cereals, fruits, vegetables, legumes, fish, olive oil, nuts and dairy products is coded as 3, consumption at least weekly as 2 and less than weekly as 1 for legumes, fish and nuts; for the other groups (cereals, fruits, vegetables, olive oil, dairy products), consumption at least daily was coded as 2 and less than daily as 1. For meat (including red meat, poultry and sausages) and dairy products, the score was partially inverted, with consumption more than weekly coded as 1, weekly as 2 and meeting the recommended consumption as 3. Moderate red wine consumption (up to 20 g/d) was coded as 3, and more or less than this daily portion was coded as 1.
After considering the different methods of calculating energy density( Reference Funtikova, Benitez-Arciniega and Gomez 18 ), we decided to present data on the basis of a dietary density calculation that includes only food items. Foods and beverages have different effects on satiety and energy intake, which in turn affects the association between energy density and body weight( Reference Ledikwe, Blanck and Khan 19 , Reference Johnson, Wilks and Lindroos 20 ). Therefore, total energy density of the diet was calculated by dividing total energy intake from food consumed each day by the total weight of the reported food intake.
Measurements were performed by a team of trained nurses and interviewers who used the same standard methods in both surveys. A precision scale of easy calibration was used for weight measurement with participants in underwear. Body weight was rounded up to the nearest 200 g and height was measured to the nearest 0·5 cm. BMI was calculated by (weight (kg)/height squared (m2)). Body weight and BMI changes were defined as the difference between the weight and BMI recorded in 2010 and at baseline in 2000, respectively.
Individuals with implausible reported energy intake (rEI) were identified by the revised Goldberg method, as described previously( Reference Mendez, Popkin and Buckland 21 ). BMR was estimated using the Mifflin equation( Reference Mifflin, St Jeor and Hill 22 ). The rEI:BMR ratio was calculated. The plausibility of rEI was estimated by comparing the rEI:BMR ratio with physical activity levels (PAL). The cut-off values to identify plausible rEI were taken as the confidence limits of agreement between rEI:BMR and PAL, and were based on the CV of participants’ energy intake, the accuracy of the BMR measurements and the total variation in PAL, as proposed by Black( Reference Black 23 ).
The validated Minnesota Leisure-Time Physical Activity (LTPA) questionnaire( Reference Elosua, Garcia and Aguilar 24 , Reference Elosua, Marrugat and Molina 25 ) was administered by a trained interviewer. Smoking habits and demographic and socio-economic variables were obtained from structured standardised questionnaires administered by trained personnel. Participants were dichotomously categorised as non-smokers (never smokers and ex-smokers with more than 1 year of smoking cessation) and current smokers (including ex-smokers with less than 1 year of smoking cessation). Maximum education level attained was elicited and dichotomously recorded for analysis as primary school v. secondary school or university.
General linear modelling procedures were used to compare baseline participant characteristics by quintiles of changes in diet cost and to analyse changes in food group consumption according to low and high changes in energy-adjusted diet cost (1st v. 5th quintile). ANOVA test and polynomial contrasts were used to determine overall P and P for linear trend, respectively, for continuous variables with normal distribution, and the Kruskal–Wallis test was used to determine overall P for non-normal distributions. P for linear trend for categorical variables was obtained by the Mantel–Haenszel linear-by-linear association χ 2 test.
Linear regression models were fitted to analyse the association between changes in energy-adjusted diet cost and changes in MDS-rec, energy density, weight and BMI. Two models were fitted. The first included three variables: sex (men/women, dichotomous), age (years, continuous) and the corresponding baseline exposure variable. The second added six variables: smoking (yes/no, dichotomous), energy intake (MJ, continuous), educational level (more than primary school yes/no, dichotomous), LTPA (metabolic equivalents,·min/d, continuous) and energy under- and over-reporting (both yes/no, dichotomous). The normality assumption of regression models was assessed by the normal probability plot. In addition, linear regression models including secular trends in diet quality as the exposure variables and changes in diet cost were fitted.
Substitution models were fitted to analyse changes in diet quality by the effect of replacing the changes in monetary costs of red meat and sausages, fast food and soft drinks, fish, cereals, dairy products and pastry with the changes in the price of vegetables and fruits. For this purpose, changes in monetary costs of vegetables and fruits were included simultaneously with red meat and sausages, fast food and soft drinks, fish, cereals, dairy products and pastry in multivariate linear regression models. The difference in the coefficients from these models was used to estimate the effect on changes in diet quality indices of replacing a 1€ increase in energy-adjusted diet costs of red meat and sausages, fast food and soft drinks, fish, cereals, dairy products and pastry with a 1€ increase in vegetables and fruits.
Cubic spline analysis was performed to investigate non-linear associations between changes in the energy-adjusted diet cost and changes in weight and BMI using the ‘gam’ package in R version 3.0.2. The assumption of normality in the regression models was assessed using the normal probability plot.
To explore effect modification according to sex, we modelled interaction terms for sex/weight change and sex/BMI change. Differences were considered significant if P<0·05. Statistical analyses were performed using SPSS version 18.0 (SPSS Inc.).
Daily diet cost increased during the follow-up by 35·1 % (online Supplementary Table S1). Substantial differences in energy-adjusted diet cost were observed between low and high diet quality at baseline and at re-examination (online Supplementary Table S1). No significant effect modification by sex was observed (P>0·1).
In the bivariate analysis, changes in energy-adjusted diet cost were positively associated with the proportion of women, age, BMI, energy consumption and energy over-reporting (online Supplementary Table S2). The opposite was true for energy under-reporting.
Differences in the changes observed in food group consumption according to a decrease (1st quintile of changes) and an increase (5th quintile of changes) in energy-adjusted dietary costs are shown in online Supplementary Fig. S2. Participants who strongly increased energy-adjusted diet cost increased their consumption of vegetables, fruits, fish and red meat and sausages and decreased the consumption of pastry, cereal products, soft drinks and fast food. The opposite was observed for those participants who decreased energy-adjusted diet cost. The strongest effect was seen for vegetables and fruits.
Diet quality increased with increasing energy-adjusted diet cost (Table 1). Changes in the MDS-rec were directly associated with increasing energy-adjusted diet costs, whereas the opposite was found for energy density (Table 1). The latter showed the strongest association with changes in energy-adjusted diet cost.
* Model 1: adjusted for sex (men/women; dichotomous), age (years; continuous) and baseline energy-adjusted diet cost.
† Model 2: model 1 plus baseline data for smoking (yes/no; dichotomous), energy intake (MJ; continuous), educational level (more than primary school, yes/no; dichotomous), leisure-time physical activity (metabolic equivalents,·min/d; continuous) and energy under- and over-reporting (both yes/no; both dichotomous).
‡ Linear regression analysis β coefficients reflect changes in energy-adjusted diet cost per 1 unit increase in continuous diet quality scores and per 1 quintile increase in categorical diet quality scores.
§ Changes in the MDS-rec.
|| Scores were standardised as a Z-value.
¶ Changes in energy density.
An increase of 1€ in energy-adjusted diet cost was associated with a decrease of 0·3 kg in body weight and 0·1 kg/m2 in BMI. These associations were no longer present when the models were adjusted for energy density (Table 2).
* Multiple linear regression analysis. β Coefficients reflect changes in body weight and BMI per 1€/8·36 MJ increase in diet cost.
† Model 1: adjusted for sex (men/women; dichotomous), age (years; continuous) and baseline scores.
‡ Model 2: includes additionally baseline data for smoking (yes/no; dichotomous), energy intake (MJ; continuous), educational level (more than primary school yes/no; dichotomous), leisure-time physical activity (metabolic equivalents,·min/d; continuous) and energy under- and over-reporting (both yes/no; dichotomous).
§ Model 3: includes additionally Δ energy density (continuous).
|| Changes in body weight.
¶ Changes in BMI.
Associations between changes in energy-adjusted diet cost and changes in weight and BMI were tested for non-linearity, but no significant evidence was found (P for curvature of changes in weight and BMI=0·47 and 0·33, respectively).
Replacing a 1€ increase in the energy-adjusted monetary cost of red meat and sausages, fast food and soft drinks, pastry and cereals with 1€ increase in vegetables and fruits significantly increased the MDS-rec (Table 3) and decreased energy density.
MDS-rec, modified Mediterranean diet score-recommended intake.
* Linear regression analysis adjusted for sex (men/women; dichotomous), age (years; continuous) and baseline data of smoking (yes/no; dichotomous), energy intake (MJ; continuous), educational level (more than primary school yes/no; dichotomous), leisure-time physical activity (metabolic equivalents, min/d; continuous) and energy under- and over-reporting (both yes/no; dichotomous). β Coefficients reflect changes in diet quality scores of replacement of 1€/8·36 MJ increased consumption of fast food and soft drinks, pastry, red meat and sausages, fish and seafood, cereals and dairy products with 1€/8·36 MJ increase in fruits and vegetables.
An increase in the energy-adjusted diet cost predicted a shift to a healthier diet and to better weight management. Diet quality strongly increased when money previously spent on unhealthy food choices such as fast food and pastry was instead spent on vegetables and fruits.
A recently published meta-analysis( Reference Rao, Afshin and Singh 5 ) concluded that healthier diets are more expensive than less healthy diets. The authors found a difference of $1·54/8·36 MJ per d between extreme quintiles of diet quality, defined by a nutrient-based dietary pattern. The monetary cost of a healthy dietary pattern, defined post hoc by cluster analysis, was twice the price of the least healthy pattern in the UK Women’s Cohort Study( Reference Morris, Hulme and Clarke 26 ). Monsivais et al.( Reference Monsivais, Rehm and Drewnowski 27 ) reported that strong adherence to the Dietary Approaches to Stop Hypertension diet was 0·78$/8·36 MJ more expensive than low adherence to this dietary pattern. In the present study, the energy-adjusted diet cost for high diet quality was 2·95€ ($3·33)/d higher than low diet quality; this amounts to 1076€ ($1215)/year for one person who chooses high diet quality. One might hypothesise that this would negatively influence healthy food choices, particularly in low-income families.
We used two conceptually different indices to measure overall diet quality: food-based and energy density, which we have shown to be good indicators of diet quality in the present population( Reference Schroder, Covas and Elosua 9 , Reference Schroder, Vila and Marrugat 10 ). Our prospective results indicate that reducing diet cost has detrimental effects on diet quality. This was true for both indicators of diet quality, underlining the robustness of our data.
In the present study, an increase in energy-adjusted diet cost of 1€ represented a 54·5 % difference between the 2nd and 4th quintiles in energy-adjusted diet cost changes. The change from a strong decrease to a strong increase in diet quality measured by adherence to the Mediterranean diet and energy density was associated with an increase of 0·42€ and 1·98€ in the energy-adjusted diet cost, respectively. For both diet quality scores, the percentage difference and percentage increase in energy-adjusted diet cost between the strong decrease and strong increase was 133 and 400 %, respectively.
The price of healthy foods increased to a greater extent than that of less healthy foods in Spain between 2000 and 2010( 8 ), and price is an important determinant for food choices( Reference French 4 ). Individuals and families facing economic constraints may be especially likely to reduce their consumption of more expensive foods, regardless of their contribution to diet quality. In addition, it is not surprising that a strong decrease in diet cost in the present study was concomitant with a dramatic decrease in the consumption of fruits and vegetables.
On the other hand, fast food and soft drinks consumption increased in participants who greatly reduced their diet cost. This is of particular concern because soft drink and fast food consumptions are associated with less healthy dietary patterns and weight management in the present population( Reference Schroder, Fito and Covas 28 ). Moreover, low diet quality is responsible for 17 % of disability-adjusted life years in the USA( 29 ). Low consumption of fruits and vegetables is one characteristic of this low diet quality. Our substitution models convincingly show the positive effect on diet quality of replacing 1€ ($0·86) increments of dietary costs in pastry and soft drinks and fast food with 1€ increases in fruits and vegetables. These data underline the paramount role of fruit and vegetable consumption in a healthy diet. Moreover, our data raise the question of food price intervention using tax policy and subsidies. Evidence indicates that a rise in prices of unhealthy foods and a price reduction for healthier alternatives improve overall diet quality( Reference French 4 , Reference Epstein, Dearing and Paluch 30 , Reference Herman, Harrison and Afifi 31 ).
Following the Mediterranean dietary pattern and low energy-dense diets have been frequently associated with better weight management and reduced risk of obesity( Reference Savage, Marini and Birch 11 , Reference Funtikova, Benitez-Arciniega and Gomez 18 , Reference Beunza, Toledo and Hu 32 ). Therefore, and based on the present results, we hypothesised that changes in diet cost would affect body weight. Our analysis showed a direct relationship between a decrease in diet cost and weight gain. This association was mainly explained by diet quality; adjusting for changes in diet quality strongly attenuated the impact of increased diet cost on weight gain.
This study has both limitations and strengths. Owing to the nature of observational studies, causal relationships cannot be drawn. Furthermore, all the dietary instruments that measure past food intake are vulnerable to random and systematic measurement errors. Although the 10-year loss to follow-up of 19·7 % in the present study can be considered acceptable, there was some evidence of selection bias among the participants who completed the follow-up in that they were generally younger and more likely to be female. Variation of monetary cost of food due to regions, seasons, and types of establishment where the food was purchased is a potential bias for the analysis of the impact of diet cost on diet quality. In the present study, we used yearly averages of food prices across multiple regions of Spain, which somewhat reduced this limitation. Furthermore, we do not have data on food consumption away from home. Our analysis was based on the assumption that most foods consumed were prepared at home. Indeed the findings of this study may not hold for those who frequently eat away from home. The strengths of the present study include its population-based design, long-term follow-up and the availability of body weight and validated lifestyle measurements at baseline and follow-up.
Results of the present study are in line with previous findings showing that healthy diets are considerably more expensive than unhealthy diets. Our prospective evidence indicates that a worsening of overall diet quality and weight development was related to a decrease in diet cost. This finding is of importance for health policy because it underlines the need to promote healthy diets that are accessible for all income levels, with implications for food pricing, agricultural and consumer subsidy programmes and tax policies.
The authors appreciate the English revision by Elaine Lilly, PhD (Writer’s First Aid).
This work was supported by grants from Instituto de Salud Carlos III FEDER (CB06/02/0029), and AGAUR (2014 SGR 240). CIBER Epidemiology and Public Health and CIBER Physiopathology of Obesity and Nutrition are an initiative of the Instituto de Salud Carlos III, Madrid, Spain.
H. S., L. S. M. and R. E. designed the research; H. S., L. S. M., I. S., M. I. P., M. F. and R. E. conducted the research; H. S. and I. S. analysed the data; and H. S. wrote the manuscript and had primary responsibility for the final content. All the authors read and approved the final version of the manuscript.
The authors declare that they have no conflicts of interest.
For supplementary material/s referred to in this article, please visit http://dx.doi.org/doi:10.1017/S0007114515005048